Results

During the time period 1989–1996, 126 803 (59.7%) of 212 448 Oregon deaths had underlying causes of death recognised by SAMMEC 3.0 as linked to smoking (table 2). Of these 126 803 deaths, SAMMEC 3.0 estimated that 42 778 (33.7%) were attributable to smoking. For the same diagnoses, physicians reported that tobacco contributed to 42 839 (33.8%) deaths, a cumulative difference of only 61 deaths over the eight year period. For each year, the ratio of SAMMEC 3.0 estimates to physician reported deaths ranged from 1.07–0.95 (fig 1). SAMMEC 3.0 estimates of deaths from neoplasms were only slightly higher than physician reported deaths from these causes; the average ratio of SAMMEC 3.0 estimates to physician reports was 1.11 with a range of 1.07–1.17. SAMMEC 3.0 generated slightly lower numbers of deaths from heart disease than those reported by physicians (average ratio 0.88, range 0.80–0.98). Deaths from respiratory disease estimated by SAMMEC 3.0 most closely matched the numbers reported by physicians, with an average ratio of 1.04 (range 1.02–1.11). Finally, although not shown by year, the overall ratio of SAMMEC 3.0 estimates of SADs to physician estimates of TADs from perinatal conditions and burn deaths was 6.82 and 2.04, respectively (table 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Annual numbers of smoking attributable deaths (SADs) estimated by SAMMEC 3.0 and physician reported tobacco attributable deaths (TADs) on death certificates, Oregon, 1989–1996. For raw data visit www.tobaccocontrol.com.

The sex and age distributions of tobacco and smoking attributable deaths reported by the two systems were also remarkably similar. Physicians reported that 61.6% of TADs occurred among men compared with SAMMEC 3.0's estimate of 62.6%. By year, physician reported deaths in men differed from SAMMEC estimates by ⩽ 3%. Overall, the median age at death for TADs reported by physicians was 73 and by year was either 72 or 73. For SAMMEC 3.0, the median age of death was 74 overall and ranged from 73–75 by year.

Lastly, we examined the 46 321 deaths during the period 1993–1996 that did not have an underlying cause of death recognised by SAMMEC 3.0 as aetiologically linked to smoking (table 3). During 1993–1996, physicians reported that 3477 (7.5%) of these deaths were related to tobacco. Of these 3477 physician reported deaths, 2227 (64.0%) listed a contributing, or secondary, cause of death that fell into one of the 27 SAMMEC 3.0 categories; this proportion was similar by year (range 62.3–65.6%). Compared with other deaths from underlying causes of death not recognised by SAMMEC 3.0 that physicians did not link to tobacco, deaths reported by physicians as related to tobacco were more than twice as likely to have a contributing cause of death aetiologically linked to tobacco. Additionally, we reviewed the underlying causes of death for the 1250 deaths that physicians reported were linked to tobacco but that did not have an underlying or contributory cause of death that fell into one of the 27 SAMMEC 3.0 categories. Four hundred and seventy one (37.1%) fell into the general category of neoplasms (International classification of diseases, ninth revision (ICD-9) codes 140–239), but no single cancer predominated. The remaining 729 deaths were widely distributed over a range of underlying causes of death, including digestive diseases (ICD-9 codes 520–579), mental disorders (ICD-9 codes 290–319), and endocrine, nutritional, and metabolic diseases and immunity disorders (ICD-9 codes 240–279).

Discussion

For the 27 SAMMEC diagnoses, SAMMEC 3.0 and physician reports provided similar estimates over the entire eight year period, suggesting that physician reporting and SAMMEC 3.0 are comparable methods of estimating smoking attributable mortality. Small discrepancies did exist: SAMMEC 3.0 estimated more deaths from neoplasms, whereas physicians reported more deaths from heart disease. Although physician estimates of deaths from lung cancer closely matched those produced by SAMMEC 3.0, they greatly underestimated SADs from cervical, oesophageal, and pancreatic cancer, suggesting a lack of knowledge of the link between smoking and these conditions. The multifactorial nature of heart disease makes it difficult to assign the specific risk factor responsible for the death on an individual basis, possibly leading physicians to overestimate the role of tobacco in this condition. Alternatively, physician reporting on death certificates may be more accurate than SAMMEC 3.0 for cardiovascular deaths; physicians are more sensitive to a patient's history of smoking behaviour and of the relation between smoking, age, and other comorbidities that are also smoking related that may have contributed to the patient's death.

SAMMEC 3.0 also estimated higher numbers of deaths from perinatal conditions and burn deaths than physicians reported. Although deaths from these conditions comprised less than 0.01% of SADs estimated by either method and did not greatly affect overall mortality estimates, it is interesting to note that physicians appear reluctant to attribute deaths from these conditions to tobacco use. It may be that physicians may not consider the role of tobacco when the deceased is not a smoker, as may happen when an infant dies of sudden infant death syndrome or the family member of a smoker dies in a fire in the home. Physicians' systematic under reporting of deaths from certain categories of deaths such as burns, perinatal conditions, and non-lung neoplasms points to the need for both medical education and media and public awareness campaigns that draw more attention to the role of tobacco in these conditions.

In some instances, physician reporting on death certificates may be more accurate than estimates derived by SAMMEC 3.0. One shortfall of SAMMEC 3.0 is its inability to consider contributing causes of death in estimating SADs. Our review of physician reported TADs from underlying causes of death not recognised by SAMMEC 3.0 found that most had a contributing cause of death that could be linked to cigarette smoking, suggesting that physician reports may result in a better estimate of smoking related mortality. Secondly, SAMMEC 3.0 estimates are based on current estimates of the prevalence of current and former smokers. After quitting smoking, the risk of death from the 27 underlying causes of death falls at different rates for the different diseases, and SAMMEC 3.0 does not take into consideration the length of time since the individual quit smoking. Physicians are in a better position to utilise cessation information, while SAMMEC 3.0 is fairly inflexible in this regard. For tracking smoking attributable mortality in settings where anti-smoking programmes have been implemented and smoking prevalence is changing rapidly, physician reporting may be more valid.

Finally, physician reporting can be a valuable source of data for risk communication to the public. Persuading legislators or the public that tobacco prevention programmes deserve public funding requires convincing evidence of the burden of smoking attributable morbidity and mortality. Estimates derived locally from trusted sources (for example, physicians) may be more useful for this purpose than from methods considered more methodologically sound by epidemiologists.